Lifetime Improvement With Predictive Maintenance of Power Electronics Based on Remaining Useful Life Prediction

This paper presents a novel predictive maintenance approach designed to enhance the longevity of power electronics components. Leveraging time series data from the NASA dataset, encompassing RDS (on-resistance) for MOSFETs and EIS (Electrochemical Impedance Spectrum) for capacitors, the proposed method utilizes Long Short-Term Memory (LSTM) neural networks for MOSFET degradation prediction and Gaussian Process Regression (GPR) for capacitor degradation. The developed LSTM-based predictive model effectively explains a significant portion of the variance in Remaining Useful Life (RUL) predictions across the dataset. Moreover, employing incremental learning on the LSTM model demonstrates remarkable adaptability and superior predictive accuracy for specific case scenarios. Additionally, the GPR model showcases its efficacy in predicting capacitor degradation. These results underscore the significance of the predictive maintenance approach in curbing unplanned downtime, extending component lifespans, and enhancing safety within power electronics systems.